test_archive.test_roundtrip_archive is flaky

[jdpage]

Specifically, it's flaky because sometimes examples generate too slowly, causing Hypothesis to fail it due to not passing health checks. Rerunning the build usually fixes it.

Tried to fix it in 154dde7 by decreasing the maximum example size, but that hasn't worked. At this point I suspect that it's due to using hypothesis' example-generation facilities in a pathological way, which is going to require digging into the docs to see if there's a better way to do it.

To make things worse, re-using the seed that hypothesis gives for reproduction in the error message doesn't always reproduce the error. However, it may result in higher error rates than random seeds, so it's probably a good place to start.